Comment by mjburgess

People repairing chatgpt replies with additional prompts is reinforcement learning training data.

"Reinforcement learning", just like any term used by AI researchers, is an extremely flexible, pseudo-psychological reskin of some pretty trivial stuff.